Tumor injection tool for mammographic scans.
Project presentation - https://www.youtube.com/watch?v=bOTE8LqXQCg&t=27s (Hebrew)
My training data set consists 2880 images augmented from 36 images of tumors marked by a specialist radiologist
Tumor injection steps:
- Crooping a square of an healthy tissue from a mammogram image.
- Resizing the square to 64x64 pixels.
- Conditioning the Image by zeroing the middel (44X44 pixel).
- Feeding the condition image to the first GAN.
- Feeding the first GAN output as input to the second GAN.
- Resizing the output to its original size.
- Adding a low power white gaussian noise for hiding the resize distortions.
- Merging the fake image into the original image using a weighted average.